2021-06-30 06:15:02

by Sasha Neftin

[permalink] [raw]
Subject: Re: [Intel-wired-lan] [PATCH] driver core: fix e1000e ltr bug

On 6/29/2021 20:33, Yee Li wrote:
>
> Yes, 18874368ns > 3145728ns.
> But, 0xe40 < 0x1003.
I got you. I would agree, direct comparison is error-prone. (10M is
impacted)
I would suggest do not use convert function. lat_en should rather
presented as lat_enc = scale x value:
Introduce two u16 variables, u16 lat_enc_d and u16 max_ltr_enc_d;

lat_enc_d = (lat_enc & 0x0x3ff) x (1U << 5*((max_ltr_enc & 0x1c00) >> 10))

max_ltr_enc_d = (max_ltr_enc & 0x0x3ff) x (1U << 5*((max_ltr_enc &
0x1c00) >> 10))

Then:
if (lat_enc_d > max_ltr_enc_d)
lat_enc = max_ltr_enc;
what do you think?

>
> So, the final lat_enc is 0xe40.
> (Latency encoded is less than maximum LTR encoded by platform)
>
> Neftin, Sasha <[email protected] <mailto:[email protected]>>
> 于 2021年6月29日周二 22:49写道:
>
> On 6/29/2021 11:21, YeeLi wrote:
> Yeeli,
> > In e1000e driver, a PCIe-like device, the max snoop/no-snoop latency
> > is the upper limit.So, directly compare the size of lat_enc and
> > max_ltr_enc is incorrect.
> >
> why?
> >      In 1000Mbps, 0x8b9 < 0x1003, 189440 ns < 3145728 ns, correct.
> >
> >      In 100Mbps, 0xc3a < 0x1003, 1900544 ns < 3145728 ns, correct.
> >
> >      In 10Mbps, 0xe40 < 0x1003, 18874368 > 3145728, incorrect.
> >
> Platform LTR encoded is 0x1003 - right. It is meant 1048576ns x 3 =
> 3145738ns.
> Now,
> for 1000M: 0x08b9 => 185ns x 1024 = 189440ns (you are correct)
> for 100M: 0x0c3a => 58ns x 32768 = 1900544ns (correct)
> for 10M: 0x0e41 => 577ns x 32768 = 18907136ns (ok?)
> 18907136ns > 3145738ns, (latency encoded is great than maximum LTR
> encoded by platform) - so, there is no point to wait more than platform
> required, and lat_enc=max_ltr_enc. It is expected and we sent right
> value to the power management controller.
> What is the problem you try solve?
>
> > Decoded the lat_enc and max_ltr_enc before compare them is necessary.
> >
> > Signed-off-by: YeeLi <[email protected]
> <mailto:[email protected]>>
> > ---
> >   drivers/net/ethernet/intel/e1000e/ich8lan.c | 23
> ++++++++++++++++++++-
> >   1 file changed, 22 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/net/ethernet/intel/e1000e/ich8lan.c
> b/drivers/net/ethernet/intel/e1000e/ich8lan.c
> > index 590ad110d383..3bff1b570b76 100644
> > --- a/drivers/net/ethernet/intel/e1000e/ich8lan.c
> > +++ b/drivers/net/ethernet/intel/e1000e/ich8lan.c
> > @@ -986,6 +986,27 @@ static s32 e1000_k1_workaround_lpt_lp(struct
> e1000_hw *hw, bool link)
> >       return ret_val;
> >   }
> >
> > +static u32 convert_e1000e_ltr_scale(u32 val)
> > +{
> > +     if (val > 5)
> > +             return 0;
> > +
> > +     return 1U << (5 * val);
> > +}
> > +
> > +static u64 decoded_ltr(u32 val)
> > +{
> > +     u64 decoded_latency;
> > +     u32 value;
> > +     u32 scale;
> > +
> > +     value = val & 0x03FF;
> > +     scale = (val & 0x1C00) >> 10;
> > +     decoded_latency = value * convert_e1000e_ltr_scale(scale);
> > +
> > +     return decoded_latency;
> > +}
> > +
> >   /**
> >    *  e1000_platform_pm_pch_lpt - Set platform power management
> values
> >    *  @hw: pointer to the HW structure
> > @@ -1059,7 +1080,7 @@ static s32 e1000_platform_pm_pch_lpt(struct
> e1000_hw *hw, bool link)
> >                                    E1000_PCI_LTR_CAP_LPT + 2,
> &max_nosnoop);
> >               max_ltr_enc = max_t(u16, max_snoop, max_nosnoop);
> >
> > -             if (lat_enc > max_ltr_enc)
> > +             if (decoded_ltr(lat_enc) > decoded_ltr(max_ltr_enc))
> >                       lat_enc = max_ltr_enc;
> >       }
> >
> >
> sasha
>